I have no idea what I'm doing, I've got this Nvidia 4060 RTX Ti, installed the CUDA drivers, and all the python scripts I run always try to allocate on the PC RAM instead of using the GPU.
4:50 I end up cutting-and-pasting the documentation into my query whenever an LLM doesn't know a new library. It's nice to have 128k/200k tokens now for GPT-4o/Claude. It's sometimes easier than reading the docs myself. Not sure if it's faster. In two years I suspect using an LLM to program will be a good skill to have.
There is no singularity. Models trained on only apples will NEVER produce oranges. The idea that Models will ever produce something they never saw is nuts. They can combine things they saw into new things as in images, stories, video, music...etc but it is just a combination of existing stuff in grammatically correct ways. Humans (Brains) on the other hand can get insights outside of themselves and use those to come up with something totally new and combine with existing things and grow knowledge. ML Models will NEVER be able to do that. These will remain useful tools to help Humans do their work. Don't believe me ? ask ChatGPT....etc for a snake game and you will see it copy / paste even variable names and their values (search online).
I migrated from the T-series to the P-series because of the P53, despite the T-series being better for use when traveling or in places other than your home or office. However, if you don't need to move the laptop and it will remain stationary, the P-series is perfect.
This is much appreciated doc. The last video on Prompts, where you suggested the "What prompt could I have used to get this output" gave me so many cool ideas. This is equally awesome.
Thanks for the very useful information. I'm currently trying to build the cheapest computer capable of running Llama3.1 8B. The NVIDIA Quadro 6Gb graphics card is the cheapest I could find so the plan is to connect two of them to be able to load the whole model. Wish me luck.
Jeff, you are the best, simply love your content! going a little deeper, I see that Keras itself comes with a Keras Tuner that offers "Bayesian" as a search model. do you have any experience with it and if yes, how do these two modules compare?
Ah so SD doesn't utilize 2 GPU's, thanks, I'm too lazy to setup my xcp-ng to give my SD VM the second RTX A5000 (which is already connected to another VM) XD
Awesome, thank you for sharing, will go through the course. If it is not too much to ask, can you please share a list of reference books, websites, etc. You recommend to augment or use as a reference for the course, thank you in advance.
No NVLink in the new generation is a very unfortunate decision by NVIDIA, now you can’t pool the VRAM from Multiple GPU‘s anymore. For my as a professional 3D Arist, who works with big and complex files, 48GB VRAM is not enough. However still good for Deep Learning
I hear the pain at 14gb... If you're like me it's not the space it takes but the wait... I think when the video cut to the time it finished you looked like you had a bit more grey in your beard.
Video not helpfull, know thy audience. Consumers with deep pockets dont need this video. So 2 x 3060 12 gb llm can utilize 2 gpu’s what can you expect?